The International Journal Of Engineering And Science (IJES)

|| Volume || 3 || Issue || 11 || Pages || PP -01-09|| 2014 ||

ISSN (e): 2319 1813 ISSN (p): 2319 1805

Comparison of Ordinary Least Square Regression and

Geographically Weighted Regression for Estimating and
Modeling the Electricity Distribution Using Geographical
Information System (GIS)/Remote Sensing (RS)
1

Kalai Selvi.J, Research Scholar, 2Vidhya .R , Professor,

3
Manonmani.R,Research Scholar
1

Institute of Remote Sensing,Anna University,Chennai,Tamil Nadu

Institute of Remote Sensing,Anna University,Chennai,Tamil Nadu
3
Institute of Remote Sesning, Anna University,Chennai,Tamil Nadu
2

-------------------------------------------------- ABSTRACT ----------------------------------------------------Electricity is a key energy source in each country and an important condition for economic development
especially in industrial area. The current energy situation in the region is characterized by a rapid increase in
energy demand due to urbanization, rapid population growth and economic growth all add to rising energy
demand . So it is very important to know the present spatial distribution of network. Understanding the
relationship between the spatial distribution of electricity network and different land use types accounting for
the spatial non stationarity can help electricity planers to better evaluate the assessment of network distribution.
A relatively new technique, geographically weighted regression (GWR) has the ability to account for spatial non
stationarity with space. While its application is growing in other scientific disciplines, the application of this
new technique in electricity distribution has not been used elsewhere. The geographic information system (GIS)
, along with the two different empirical techniques( GWR and Ordinary least square regression) was used to
analyze the relationship between low tension (LT) distribution and various land use classes derived from recent
high resolution satellite image quick bird for Manali (Industrial region) in Chennai. Low tension was spatially
interpolated in ArcGIS using interpolation techniques with zonal statistics. The explanatory variables used are
the Land use parameters like built-up area, scrub, agricultural land, industry etc and the socio economic factor
population growth. The OLS model performed moderately well (AIC=31.665, R2=31.9% and Adjusted
R2=31.8%), the Morans I =0.66 for the residuals from the OLS model. The best results were obtained with the
GWR model (AIC=19.08, R2=51.65% and Adjusted R2=50.42%) The results suggest that GWR provides an
effective estimation for modeling the LT consumers distribution network pattern.

KEYWORDS: Remote Sensing, Geographical Information System, Geographically Weighted Regression,

Ordinary least square component;
--------------------------------------------------------------------------------------------------------------------------------------Date of Submission: 27 October 2014
Date of Accepted: 15 November 2014
---------------------------------------------------------------------------------------------------------------------------------------

I.

INTRODUCTION

Energy has come to be known as a `strategic commodity and any uncertainty about its supply can
threaten the functioning of the economy, particularly in developing country like India. As demands for the
electricity energy have increases in urban area due to tremdeous population growth as a result in change in land
use patterns and industry establishment. Energy is critical, directly or indirectly, in the complete process of
evolution, growth and survival of all living beings and it plays an important role in the socio-economic
development and human welfare of a country . Hence, to offer high-quality service to the end user considering
the urbanization growth an adequate planning for energy or demand estimation is very much essential.
In recent years Geographic information system has been an important development in the field of
electricity network [1][2] and it is used for the spatial data management and manipulation. With the advent of
remote sensing and GIS technologies, the mapping of electricity distribution network with considering socio
economic and land use variable have been widely used[3]. There are many studies on electricity related to trend
analysis[4] and many these studies applied regression method as non spatial[5] .
Recent new technology known as Geographically weighted regression is applied to study the spatial
relationship between more than two variable. One of the nonparametric modeling method is the geographically
weighted regression (GWR) technique[6]. GWR is among the new developments of local spatial analytical

www.theijes.com

The IJES

Page 1

Comparison of Ordinary Least Square Regression and Geographically Weighted Regression .

techniques. It is a local spatial statistical technique that relies on a form of kernel regression within a multiple
linear regression framework to develop local relationships between the dependent and independent
variables[7][8][9]. GWR was used to examine the spatially varying relationships between several urbanization
indicators based on LULC changes. Geographically weighted regression is an exploratory technique mainly
intended to indicate where non-stationarity is taking place on the map, that is where locally weighted regression
coefficients move away from their global values. In another research[10] the relationship between precipitation
verse irrigated and rain fed crop was carried out. Another research [11] is about analysis the relationship
between agricultural landscape pattern and urban.
Numerous studies have described the application of the Geographical weighted regression method for
land use change. In addition, the geographical weighted regression method has been applied to urban heat
estimation [12], urban growth [13], crime mapping [14], Fisheries [15], population [16], Electricity
consumption[17] and Groundwater subsidence [18]. Further GWR was used to investigated relationship
between electricity consumption and household income. The results reveal that electricity consumption is useful
for characterizing household income, a frequently used proxy for purchasing power. Stochastic model to
estimate the efficiency analysis of electricity distribution[19] network using sample of about 500 electricity
distribution utilities from seven European countries. Although Geographical regression method has been applied
to crime mapping, fisheries, land use mapping, heat island, this approach have not yet been used in analysis
relationship between distribution of low tension and land use type.
At present, energy modeling is a subject of widespread interest among engineers and scientists
concerned with the problems of energy production and consumption. Modeling in some areas of application is
now capable of making useful contributions to planning and policy formulation. GWR and OLS have the
ability to predict the locations where the network expansion will occur is useful not only for proper utilization of
the power , but also for policy makers who need to plan and manage the outcomes of spatial processes at
regional or local levels. Spatially estimating the future demand is very important for the economical future
expansion and safe operation of a distribution network
This study compares the accuracy in predicting energy consumption in the study area using OLS and
GWR and also examines the relationship between energy consumption and diverse independent variables. The
energy consumption estimation is calculated based on Land use parameters and also based on population.
Results are compared and this study provides important reference materials for the utility companies in
assessing energy consumption.
This will be helpful for a successful electricity planner.

II.

AREA AND METHODS FOR RESEARCH

a. Objective
The main objective of this study includes
Applying OLS and GWR to estimate the distribution pattern of LT Network for planning future expansion
based on the independent variables.
Evaluating the performance of OLS and GWR models.
Validation of the results.

b. Area of research
The study area selected is situated in the city Chennai in the state of Tamil Nadu.The Chennai
Metropolitan Area (CMA) comprises the city of Chennai, 16 Municipalities, 20 Town Panchayats and 214
Village Panchayats in 10 Panchayat Unions. Manali is an Industrial town and Municipality in Thiruvallur
district in the state of Tamil Nadu. It is located in north of Chennai City . Manali is further divided into six
wards and Manali had a population of 58,174 as per the census data. Manali is located in the Northern Suburb
of Chennai city. Since this area consist both industrial as well as domestic loads this has been taken for case
study and its the one of the fast developing areas in Chennai whereas the load growth is in the rate of around
8% per year. Manali area is fed by the Manali 230/110KV Substation which in turn connected with the thermal
generators located in north chennai. The study area Manali is shown in Figure[1].

www.theijes.com

The IJES

Page 2

Comparison of Ordinary Least Square Regression and Geographically Weighted Regression .

Fig.:1 Study area

III.

OVERVIEW OF MODELS

A. Ordinary least squares (OLS) method

Various interpolation techniques are available to predict and interpolate information or variables within
predetermined boundaries. The ordinary linear regression model the estimation of the coefficients by using
Ordinary Least Squares. The residuals are assumed to be independently and identically distributed around a
mean of zero. The errors are also assumed to be homoscedastic, i.e with constant variance. A regression model
is expressed as inn equation(1).
y i 0 1 x i i -----------(1)
for i=1....n
Where Y is the dependent variable. The independent variables are known as predictor variables. The i is the
error term, and 0 and 1 are parameters which are to be estimated. The OLS estimator can be written in the
form shown in equation(2)
( X

where

X )

---------------- (2)

is the vector of estimated parameters, X is the design matrix which contains the values of the
T

independent variables and a column of 1s, y is the vector of observed values, and ( X X ) is the inverse of
the variance-covariance matrix. Weights can also be included in the OLS estimator and they are placed in the
leading diagonal of a square matrix W , the estimator with weights are shown in equation(3).
( X WX )
T

Wy ---------- (3)
The ability of the model to replicate the observed y values is measured by the goodness of fit. This is expressed
by the r2 value which runs from 0 to 1 and measures the proportion of variation in the observed y.
X

B. Spatial Autocorrelation
Autocorrelation means that a variable is correlated with itself . The simplest definition of
autocorrelation states that pairs of subjects that are close to each other are more likely to have more similar
values , and pairs of subjects far apart from each other are more likely to have less similar values . Gradients or
clusters are examples of spatial structures that are positively correlated, whereas negative
correlation may be exhibited in a checkerboard pattern where subjects appear to repulse each other. When data
are spatially auto correlated, it is possible to predict the value at one location based on the value sampled from a
nearby location when data using interpolation methods. The absence of autocorrelation implies data are
independent. Moran's I is a measure of spatial autocorrelation.
C. Geographically Weighted Regression (GWR)
GWR is the analysis of spatially varying relationship .GWR is to explore how the relationship between
a dependent variable (Y) and one or more independent variables (the Xs) might vary geographically.
Geographically Weighted Regression (GWR) is a recent contribution to modeling spatially heterogeneous
processes. Using GWR the parameters may be estimated anywhere in the study area by the given dependent
variable and a set of one or more independent variables which have been measured at places whose location
coordinates are known. GWR model considers the differences of spatial location and the spatial correlation,
which allows local rather than global parameter estimation, the estimated parameters are different with the
spatial location varies. It can be regarded as a local model .

www.theijes.com

The IJES

Page 3

Comparison of Ordinary Least Square Regression and Geographically Weighted Regression .

The GWR model equation would be:
y i ( u ) 0 i ( u ) 1 i ( u ) x 1 i 2 i ( u ) x 2 i ... mi ( u ) x mi ---------------- (4)

A prediction/estimation may be made for the dependent variable if measurements for the independent
variables are also available at the same location u. In general the GWR works by moving a search window from
one point in a data set to the other, working through them all in a ordered list or sequence. The distance from
one point to another can be defined by actual geographic distance or by its sequence of position i.e first nearest
point or second and so on. The goodness of fit measured in GWR is the corrected Akaike Information Criterion.
AIC

2 n log

( ) n log

n tr ( S )
( 2 ) n
n 2 tr ( S )

------------------(5)

In AIC method, the user can choose a fixed bandwidth or a variable bandwidth that expands in areas of
sparse observations and shrinks in areas of dense observations (Charlton et al., no date). Because the regression
equation is calibrated independently for each observation, a separate parameter estimate, t-value, and goodnessof-fit is calculated for each observation. These values can thus be mapped, allowing the analyst to visually
interpret the spatial distribution of the nature and strength of the relationships among explanatory and dependent
variables.

IV.

MODEL PARAMETER

A. Dependent variable
The feeder emanating from the substation has been mapped in GIS along with all roads and buildings.
A very high resolution map taken from satellite is used us to map network elements in GIS and the spatial coordinates of poles ,transformers, individual service lines from the pole etc has been acquired. Further, the
attribute data like make of distribution transformer, capacity of each DT, source feeder details, year of
commissioning etc are also recorded along with each object which is given in Figure[3&4].
B. Independent variable
Further, land-uses of Manali area has been classified into nine categories. Viz. built up, canal, crop,
Cooling pond, Industry, Plantation, Scrub, River and Tank. In this study supervised classification techniques
were adopted. A supervised classification method was carried out using training sets and test data for accuracy
assessment. Classified land use / land cover maps for the year 2006 and 2012 . After classified thematic maps
were developed, accuracy was tested by different methods of accuracy assessment, and the post-classification
process was the last process in classification. This was considered as input for the model.
C. Demographic profile
Population growth in the study area
Tamilnadu has emerged as the third largest economy in India.

Fig.: 2 Chennai Population till 2014.

In the recent past, liberalization, rapidly growing IT sector, an educated, hardworking and disciplined
work force etc, accelerating economic development also contributed to the growth of urban areas in Tamilnadu.
The extent of the State is130,058 sq.km. of which the urban area accounts for 12,525 sq.km. Tamilnadu is the
most urbanized state in India. Chennai was established in 1639 and it has grown to the fourth largest Metro City
in India.Fig.2 shows the population growth is Chennai. According to India Census , the study area Manali had a
population of 58,174. Manali has an average literacy rate of 72 %, higher than the national average of 59.5 %
.The population growth will also play an important role in energy consumption. Hence this also been considered
in the analysis.

www.theijes.com

The IJES

Page 4

Comparison of Ordinary Least Square Regression and Geographically Weighted Regression .

Fig.:3&4 Mapping of Distribution Network

V.

RESULT AND DISCUSSION

The relationship between the electricity consumption pattern with the independent factors viz. Land use
parameters and population growth the socio- economic factor are estimated using OLS regression model, auto
correlation and GWR model. The land use map and the corresponding estimation results observed using OLS
Regression model is shown in Fig(5) and (6). It is observed from the result that the spatial estimate result of LT
Consumers using OLS model falls maximum under the built-up area and it is very minimum in scrub land.
Table 1.
R Values from OLS MODEL
Object
Id
1
2
3
4
5
6
7
8

Diag_Name
AIC
R
AdjR
F-Stat
Wald
K(BP)
JB
Sigma

Diag_Value
31665.68
0.319
0.318
674.759
305.241
1587.229
259669.648
14.076

Fig.: 5. Land use and Land cover

www.theijes.com

The IJES

Page 5

Comparison of Ordinary Least Square Regression and Geographically Weighted Regression .

Fig.:6.OLS Model Regression

Under the built-up area and it is very minimum in scrub land. The R values obtained from OLS is
displayed in Table [1]. The t-statistics values observed from the coefficient table are shown in Table[2].
Table-2.
Coefficient Table
Id

Coef

StdError

t_Stat

Prob

0.021042

0.068954

0.305164

0.760262

0.102445

0.002004

51.11694

0.003086

0.001363

2.264861

0.023542

0.014374

0.011825

1.215536

0.224214

-0.00157

0.038385

-0.04089

0.967373

The t-statistics test the hypothesis that the value of an individual coefficient estimate is not significantly
different from zero. Here for the variable Built-up area, t-statistics values are more statistically significant.

Fig.:7 t-statistics Graph

The report from the OLS advised that we should carry out a test to determine whether there is spatial
autocorrelation in the residuals. If the residuals are sufficiently auto correlated, then the results of the OLS
regression analysis are unreliable. An appropriate test statistic is Morans I, this is a measure of the level of
spatial autocorrelation in the residuals. Auto correlation measures spatial autocorrelation based on both feature
locations and feature values simultaneously. It should be between -1.0 to 1.0 . The results of auto correlation
estimates are shown in Table.[3]

www.theijes.com

The IJES

Page 6

Comparison of Ordinary Least Square Regression and Geographically Weighted Regression .

Table : 3 Auto correlation result
Morans Index : 0.66
Z score : 53.61 standard Deviation
Significance Level

Critical Value

0.01

-2.58

0.05

-1.96

0.10

-1.65

Random

-1.65 to 1.65

0.10

1.65

0.05

1.96

0.01

2.58

With the given a set of features and an associated attribute, it evaluates whether the pattern expressed is
clustered, dispersed, or random. The calculated the Moran's I Index value and both a z-score and p-value to
evaluate the significance of that Index. P-values are numerical approximations of the area under the curve for a
known distribution, limited by the test statistic. To estimate the location of LT consumers after auto correlation
the GWR model was used to identify the hotspots and the local results R is shown in Fig.[8].
The hotspots in red colored squares Fig.[9] indicate the hotspot areas where the electricity consumers
are more and their total consumption is also high compared with the other regions. Even the area around this
hotspot is also built up area the reason for this estimation is, this area contains multi-storey buildings with more
than one connection for an individual building. The Resultant table obtained from GWR is given as Fig.[4].
Table 4 : GWR Resultant Table.
Name
Bandwidth
Residual squares

Value
895.430243
55865.531082

Description

Effective Number

80.742341

Dependent variableLT Consumer

Sigma
AIC
R
R Adjusted

4.129999
19084.301630
0.516006
0.504222

The GWR table is for measuring the goodness of fit. It contains Residual squares , r 2 , adjusted r2 and
the sigma values. The r 2 measures the proportion of the variation in the dependent variable which is accounted
for by the variation in the model, and the possible values range from 0 to 1. Values closer to 1 indicate that the
model has a better predictive performance. However, its values can be influenced by the number of the variables
which are in the model increasing the number of variables will never decrease the r2. The adjusted r 2 is a
preferable measure since it contains some adjustment for the number of variables in the model. Goodness of fit
measurements : the r2 is 0.516 and the adjusted r 2 is 0.504. The comparative table is given as Table[5].

NAME
AIC
R
R
ADJUSTED
SIGMA
SIGMA

www.theijes.com

TABLE 5: VALIDATION OF MODELS

VALUES FROM
VALUES FROM
GWR
OLS
19084.301630
31665.680
0.516006
0.319
0.504222
0.318
4.12999
14.076

_
_

The IJES

Page 7

Comparison of Ordinary Least Square Regression and Geographically Weighted Regression .

Fig. 8: GWR Local R

Fig. [9] Geographically weighted Regression predicted results

It is inferred from the above the results that the R values close to the value 1 indicates better
estimation results and the lowest Akaike Information Criterion (AIC) value which is the relative measure of
goodness-of-fit indicates better results. From the result the OLS model performed moderately well
(AIC=31.665, R2=31.9% and
Adjusted R2=31.8%), the Morans I =0.66 for the residuals from the OLS model. The best results were
obtained with the GWR model (AIC=19.08, R2=51.65% and Adjusted R2=50.42%) Wherein in GWR model
gives better results of estimation than OLS model. The estimated values are compared with the actual physical
distribution of network.

VI.

CONCLUSION

This study explored the use of ordinary least squares (OLS) regression, spatial autocorrelation and
geographically weighted regression (GWR) for modeling and analyzing the spatial varying relationships between
LT Consumers and land use pattern in the study area. Results lead to the conclusion that GWR was more
powerful and effective in interpreting relationships between LT consumer and land use pattern, particularly in
relation to urbanization. Characters and strength of the relationships identified by GWR showed great spatial
estimates. Given that impacts of different urbanization indictors of landscape patterns operated at different spatial
scales, the OLS and GWR estimated the dependent variables distribution pattern from other independent variables
or drivers. This study can be preceded further by using both the regression models for two different type of study
areas and results can be analyzed based on its urbanization growth rate.

www.theijes.com

The IJES

Page 8

Comparison of Ordinary Least Square Regression and Geographically Weighted Regression .

ACKNOWLEDGMENT
The author would like to thank TamilNadu Transmission Corporation Ltd., Chennai for providing data for this
research work.

REFERENCES
[1].
[2].
[3].
[4].
[5].
[6].
[7].
[8].
[9].
[10].

[11].
[12].
[13].

[14].
[15].

[16].
[17].

[18].
[19].

A. Nagaraha Sekhar, K.S Rajan and Amit Jain, Application of Geographical Information system and spatial informatics to
Electric power systems Fifteenth national power system conference ,NPSC,IIT Bombay, 2008
T.Jamal, , W. Ongsakul , M.S.H Lipu and M.S, Islam An approach to integrate Geographic information system to the proposed
smart grid for Dhaka, Bangladesh, International journal of scientific engineering and technology, vol No 3, PP124-129,2014
J. Kalai selvi. R.Vidhya R. Manonmani, Geospatial technique in urban distribution network mapping and modeling International
journal of Asia Pacific research, vol 1, pp 63-70,2014
P. Chujai, N Kerdprasop and K. Kerdprasop , Time series analysis of household electric consumption with ARIMA and ARIMA
models Proceedigs of the international Multi conference of engineers and computer scientists, vol IMECS 2013, 2013
V. Lepojevic and M. Andelkovic Pesic Forecasting electricity consumption by using holt-winters and seasonal regression
models Facta Universitatis Economics and Organization vol 8, no 4,pp421-431,2011
M.Cahill ,G. Mulligan .Using geographically weighted regression to explore local crime patterns journal Soc Sci Comput Rev ,
25:174193,2007
Fotheringham, A.S., Charlton, M. & Brunsdon, C., 1997a; Measuring spatial variations in relationships with geographically
weighted regression, in Fischer, M.M. & Getis, A. (eds), pp 60-82. 1997
Brunsdon, C., Fotheringham, A.S. & Charlton, M.E., 1996; Geographically weighted regression - a method for exploring spatial
nonstationarity, Geographical Analysis, 28 (4), 281-298. 1996
Robinson, D.P., Lloyd, C.D., McKinley, J.M. Increasing the accuracy of nitrogen dioxide (NO2) pollution mapping using
geographically weighted regression (GWR) and geostatistics. Int. J. Appl. Earth Obs. 21, 374-383,2013
V. Sharma, S. Irmak, I. Kabenge Application of GIS and GWR to evaluate the spatial non-stationarity relationship between
precipitation Vs irrigated and rain fed maize and soybean yields American society of Agricultural and biological engineers , vol
54(3) pp 953-972,2011
S. Su , R. Xiao and Y. Zhang 2012, Multi-scale analysis of spatially varying relationships between agricultural landscape patterns
and urbanization using geographically weighted regression, Applied Geography 32 ,2012
M. Szymanowski and M. Kryza,2012 Application of geographically weighted regression for modeling the spatial structure of urban
heat island in the city of Wroclaw (SW Poland) Procedia Environmental Sciences 3 pp 1419,2011
N.M Shariff , S. Gairola and A. Talib Modeling Urban Land Use Change Using Geographically Weighted Regression and the
Implications for Sustainable Environmental Planning, International Environmental Modeling and Software Society (iEMSs) 2010
International Congress on Environmental Modeling and Software Modeling for Environments Sake, Fifth Biennial Meeting,
Ottawa, Canada,2010
J. Lu , T .Guo-an The Spatial Distribution Cause Analysis of Theft, Crime Rate Based on GWR Model, Multimedia Technology
(ICMT), 2011 International Conference 26-28 July 2011, Hangzhou ,IEEE,2011
Matthew J. S. Windle, George A. Rose, Rodolphe Devillers and Marie-Jose Fortin Exploring spatial non-stationarity of fisheries
survey data using geographically weighted regression (GWR): an example from the Northwest Atlantic ,ICES journal of Marine
science vol 67, pp 145-154,2009
Eduardo Francisco (2008) , A Consumer Income Predicting Model Based on Survey Data: An Analysis Using Geographically
Weighted Regression (GWR), Latin American Advances in Consumer Research, vol 2, pp 77-83
S.Rong-Kang ,S. Yi-Shiang , M.Kuo-Chen Using geographically weighted regression to explore the spatially varying
relationship between land subsidence and groundwater level variations: A case study in the Choshuichi alluvial fan, Taiwan IEEE,
2011
Christian Growitsch, Tooraj Jamasb and Michael Pollitt Quality of Service, Efficiency and Scale in Network Industries: An
analysis of European electricity distribution, IWH-Diskussionspapiere ,2005
An Introduction to Spatial Autocorrelation Analysis with GeoDa Luc Anselin Spatial Analysis Laboratory Department of
Agricultural and Consumer Economics University of Illinois, Urbana-Champaign.June 16, 2003

www.theijes.com

The IJES

Page 9